I have big plans for this page, but I haven't taken the time to build anything for you yet. Eventually I intend to put some of my own software online here. In the meantime, why not play a game of 20 Questions against an AI?

*** How it Works***

20Q is a neural network that uses the pattern of your responses to "read your mind". It learns these patterns simply by examining data from millions of previous games. You'll be amazed at how well it can figure out what you're thinking!

20Q works on the principle of using each question to reduce information entropy. All this really means is that the computer is trying to eliminate as many options as possible with each question it asks you. If each question could eliminate half of the possibilities, then the computer could use 19 questions to distinguish between 524,288 (or 2^19) different possible answers. The Oxford dictionary only has 171,476 words!

So to play 20Q your computer needs a list of all possible words along with a database of answers about those words (is it alive? do you find it in a house?). Then it chooses which question to ask next by checking the database to see which answer will eliminate about half of the remaining possibilities. But that's not very "intelligent"; where does the AI come in?

Computers are generally bad at dealing with human error or ambiguity. Many people might say "yes" when asked if a mushroom is a plant -- even though a mushroom is actually a fungus. This could cause the computer to unfairly eliminate "mushroom" from the set of possible answers. A person is bound to give at least one ambiguous answer during a game of 20 questions... so this is where we need the AI!

Neural networks automatically learn to find patterns in information. If you ask 100 people to answer a question about mushrooms, maybe 99 will say "yes, they are found in the woods" while only 50 might say "no, it is not a plant". The neural network takes these numbers and, using only mathematics, deduces which pattern of answers is really important in determining if something is a mushroom. It will learn to ignore the question about plants and put trust into questions like "is it a food?" and "does it grow?". If the general pattern of answers is mushroom-like even though some answers seem to disagree, the neural network can still guess mushroom. This doesn't mean the computer is intelligent, just less dumb. In a nutshell, acting less dumb is what AI is all about!